Reviewed: tsc + consult 295 + rectification 747 + python suites green;
guard functions and window/general instructions verified unchanged.
Conflict: BUG entry renumbered 467→468 (467 taken by report-extraction).
Review fix: test_consultation_birth_accuracy.py added to CORE_PYTEST_TARGETS
(pytest-style file outside the rectification glob would never hit the auto gate).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0155nFCgCHtoA7jhSDGmZmMu
The report writer had no interpretation methodology at all: a local agent
calling the jyotish skill can read the reference library, the report model
could read nothing. It could only restate the bundle.
- frontend/src/lib/report-interpretation-packs/ holds one general pack and
one pack per report theme, distilled from the in-repo reference guides.
They constrain wording and reasoning discipline (term modernisation,
how to talk about relative strength and SAV scores, the reasoning errors
to avoid, the banned phrasings) and never assert a chart fact.
- The general pack rides INSIDE the cached system message so the cached
prefix stays byte-identical across sections; the chapter pack follows it
and summary calls get the general pack only.
- Skill jyotish-personal-report goes to 1.1.0 (1.0.0 deprecated): the
contract now names interpretiveFacts and themeNarrativeSeeds as a
bounded fact layer and states that the knowledge pack is not a fact
source and cannot raise certainty.
- Telemetry records interpretiveFactCount and knowledgePackCharacters as
numbers only; the counter never throws so telemetry cannot break a run.
Tests lock every theme resolving a pack, the 3,000 character budget, a
forbidden-substring scan (paths, module names, vendor names, artefact
names), the byte-stable cache prefix, and that no evidence id or date
appears in the static content.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016P5RoqzmUQEbeC2qjAkeGr
The consultation workflow already returns a functional benefic/malefic
table, a shadbala ranking, SAV scores, the current maha/antardasha,
detected yogas and guided-topic copy. The report extraction layer threw
all of it away, so claim cards could only say "the server closed the
minimum evidence group" and the writer had no conclusions to work from.
- ReportEvidenceBundleV2 gains interpretiveFacts (yogas, functionalRoles,
shadbalaRanking, savScores/savTotal, currentDasha, convergenceDomains)
and themeNarrativeSeeds. Both are required, allow empty, keep .strict(),
are covered by the canonical sort + bundleHash, and fail closed on
dangling refs, duplicate ranks/houses/themes and out-of-bound text.
- Extraction is allowlist-style: closed enums for yoga category and
functional role, safeCelestialName for planets, sign->whole-sign-house
projection for SAV, and a forbidden-token scrub that drops any seed line
naming an external provider or internal route.
- Claim card conclusions and supportingFacts are now deterministic
astrological statements built from those facts; risks become
counterFacts. assertionLevel derivation is unchanged, and a theme with
no seed keeps the old receipt wording with consensus capped down.
- filterReportEvidenceBundleForSection trims seeds and SAV houses to the
chapter's theme while letting the chart-wide interpretive receipts ride
along, so every section can cite them.
Contract snapshot taken from a real local /api/consultation_workflow call
with fictional smoke birth data; the new fixture test locks the shapes
that call actually returns, including the fields that are absent.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016P5RoqzmUQEbeC2qjAkeGr
Natal answers were opening on parameter tables, and rectification turns were one-sentence legal copy. Centralize user-facing strings, keep representative-minute and question-slot red lines, and stop duplicating the opening collect prompt as a second assistant message.
Co-authored-by: Cursor <cursoragent@cursor.com>
Hospital records and adopted rectification times were still fed to the model as not_auto_rectified because the chart request omitted declared_accuracy/time_source and mastra hardcoded the boundary. Map profile truth into the engine request, keep rectified for accepted/confirmed active times only, and leave window/general guards unchanged.
Co-authored-by: Cursor <cursoragent@cursor.com>
Window_scan assertions now match the public from_sign/to_sign contract, the
staging quick gate runs the rectification Python suite, and compare-candidates
rescores when stored policy lags the live engine identity.
Co-authored-by: Cursor <cursoragent@cursor.com>
Choice path already wrote spokenFollowupForUser into the body; free-text
dropped that stem and only filled empty answers, so a new collect_spoken
focus stayed invisible after “记下了”.
Co-authored-by: Cursor <cursoragent@cursor.com>
The standalone collect prompt bar reused the choice-card chrome and
duplicated the question. Visibility now comes from the spoken reply,
with an empty-body fallback that posts the persisted focus prompt.
Co-authored-by: Cursor <cursoragent@cursor.com>
Occupation answers were stored as career, so coverage never closed, yearless
cards never fired, and the decision layer suppressed an engine-allowed range.
Normalize occupation-collect writes, allow yearless cards once training is
open, offer a range when no distinguish card remains, and align the public
house table to the representative minute.
Co-authored-by: Cursor <cursoragent@cursor.com>
Spoken collect prompts lived only in GET current_question. The chat never
parsed that field, Agent projections returned null after evidence writes,
and active_focus followups collapsed the questionId. Render the parsed
prompt, keep choiceReady on real cards, and give collect focuses a stable
domain-scoped id.
Co-authored-by: Cursor <cursoragent@cursor.com>
The choice and collect fast paths applied the answer then returned,
so a dated event in the same utterance never reached the evidence
ledger. Idle persist also prechecked follow-up with a hardcoded
collect_evidence outcome instead of the dossier decision.
Co-authored-by: Cursor <cursoragent@cursor.com>
A spoken no neither scored nor declined coverage, so relatives never closed and free-text turns left current_question null. Prefer a same-domain yearless scoring card before an unscoreable collect, resolve explicit collect denials as declined, and persist the next followup after an idle agent turn.
Co-authored-by: Cursor <cursoragent@cursor.com>
Scoring scoped the contrast packet to the previous round's candidate set while
minting probes for the new one, so every later read re-prefixed the stored hash
and the persisted focus schema could never match. Mint one prefix per set and
compare splits by probe identity.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LVapmh2oGNyr6ECHKjPJY8
Collect focus was leaking discriminator probes into the agent prompt and
inheriting choice identity. Mark spoken questions, hide current_probe
unless a real choice card exists, strip fallback identity, log persist
failures, and align the decision probe with the plan including birthDate.
Co-authored-by: Cursor <cursoragent@cursor.com>
Silent unrenderable discriminators, a missing question-contract golden, and a always-on tool table were hiding fail-closed drops behind the prompt wall.
Co-authored-by: Cursor <cursoragent@cursor.com>
Missing evidence fingerprints are stale, not current. Stale snapshots keep asking discriminators when a probe exists, and offer-candidates refuses until the ledger matches.
Co-authored-by: Cursor <cursoragent@cursor.com>
Reserve a month-or-better holdout, degrade unsigned D9/D10 to existence, and treat ledger keywords as mention-only so remaining discriminators stay in the pool.
Co-authored-by: Cursor <cursoragent@cursor.com>
Closing a discriminator used to leave GET without a card after refresh.
Write the next dated question in the same request, skip childhood career
and move probes, and do not continue a read-only turn when that question
is already persisted.
Co-authored-by: Cursor <cursoragent@cursor.com>
Family and occupation method layers were blocking discrimination even when
training events were complete and a discriminator probe existed, so the agent
only acknowledged evidence and stopped.
Co-authored-by: Cursor <cursoragent@cursor.com>
Docker next build failed on ProbeAnswer.id and optional decisionReceipt,
so staging never received the month-lock SHA.
Co-authored-by: Cursor <cursoragent@cursor.com>
Dated evidence in a domain no longer yields another existence question
in that domain. Remaining varga discriminators all stay in the pool so
the next card is whichever unused split scores highest.
Co-authored-by: Cursor <cursoragent@cursor.com>
Empty snapshot candidates were starving remaining D24 splits, so the
TypeScript follow-up chain asked the low-gain Python career probe.
Read paths now share one inference+engine catalog and yield a stale
low-gain distinguish card to the current winner.
Co-authored-by: Cursor <cursoragent@cursor.com>
Python and TypeScript now share a four-option probe contract, persist Focus before asking, and pick the highest-value renderable probe instead of preferring low-gain career events over D24.
Co-authored-by: Cursor <cursoragent@cursor.com>
Three collected events with a reserved holdout were stalling because the discriminator door counted holdout. Public selection_allowed still had snapshot fallbacks, and health only proved the image SHA.
Co-authored-by: Cursor <cursoragent@cursor.com>
Showing a choice card is no longer treated as completion. Distinguish probes
require real candidate groups, holdout stays out of scoring, and ordinary
sessions can finish with a credible range instead of an exact-minute gate.
Co-authored-by: Cursor <cursoragent@cursor.com>
Exam-quality cards may still jump ahead of adoption, but career years stay on method rotation. Server stamps only period and family; spoken questions remain model-authored.
Co-authored-by: Cursor <cursoragent@cursor.com>
Recorded-year quality probes were spoken-only, so the interview had no
choice card. Compare also re-scored after batch until the 105s attempt
aborted the turn.
Co-authored-by: Cursor <cursoragent@cursor.com>
Conflict probes were jumping after one dated event, so the interview asked
another domain before method collection. Spoken replies now follow the
stamped choice prompt instead of a topic denylist.
Co-authored-by: Cursor <cursoragent@cursor.com>
Evidence writes now return the persisted open_question so the model asks that stem instead of a second education probe, and the jump-to-latest chip is centered again.
Co-authored-by: Cursor <cursoragent@cursor.com>
Choice cards used a hardcoded domain menu and always asked existence.
Rank scoring layers by remaining-minute entropy, keep finance and health
volunteer-only, and ask D9/D10 style or exam quality so taps match outcomes.
Co-authored-by: Cursor <cursoragent@cursor.com>
Coverage-complete ties stayed in discrimination because whole-window D9/D24 follow-ups were treated as probes, and restated dates inserted duplicate evidence. Skip encoded remaining layers, ask leftover D4 or offer a provisional range, and dedupe dated rows by kind and date.
Co-authored-by: Cursor <cursoragent@cursor.com>
Give the interview a hidden reasoning channel so planning leaves the spoken reply, publish terminal text-delta as-is, and stop regex or Case templates from replacing the model.
Co-authored-by: Cursor <cursoragent@cursor.com>
Coverage-complete ties never persisted A/B/C/D because contrast probes were stamped with an answered education quality probe, remaining minutes were asked as window D10 signs, and 「没有了」 missed the stop pattern.
Co-authored-by: Cursor <cursoragent@cursor.com>
Mastra intermediate text-delta was published as answer.delta, then set-focus domain errors reset the attempt and replayed evidence. Publish only the terminal no-tool step, persist the next probe on the server, and ground batch quotes in the source turn.
Co-authored-by: Cursor <cursoragent@cursor.com>
Clicking A/B/C/D or stop must persist the answer, close the probe, and
update posteriors in one idempotent transaction instead of sending the
option text as a chat message.
Co-authored-by: Cursor <cursoragent@cursor.com>
Coverage complete only unlocks discrimination. A 34/33/33 window plus an
occupation note must ask a D9/D10 contrast probe instead of offering a
stale winner card.
Co-authored-by: Cursor <cursoragent@cursor.com>
Jesse_Chen